Mammography Image Databases

The Digital Database for Screening Mammography (DDSM) is supported through a grant from the DOD Breast Cancer Research Program, US Army Research and Material Command DAMD17-94-J-4015. The server for the DDSM is a dual processor Sun Sparc 20 with 520 Megs of RAM donated by Sun Microsystems through their Academic Equipment Grant (AEG) program, Grant #: EDUD-US-950408.


o Digital Database for Screening Mammography:

A project to establish a database for use by the mammographic image analysis research community has been awarded a grant through the Breast Cancer Research program of the U.S. Army Medical Research and Development Command. This is a collaborative effort involving Massachusetts General Hospital, the University of South Florida, and Sandia National Laboratories. The primary purpose of the database is to facilitate sound research in the development of computer algorithms to aid in screening. Secondary purposes of the database may include the development of algorithms to aid in the diagnosis and the development of teaching or training aids. The database will contain approximately 3,000 studies. Each study includes two images of each breast, along with some associated patient history information and image information. Images containing suspicious areas will have associated pixel-level "ground truth" information about the locations and types of suspicious regions. Also provided will be software source code both for accessing the mammogram and truth images and for calculating performance figures for automated image analysis algorithms.

Volume 1 and Volume 2 of the database are now available. Each contains over 100 normal cases. A special volume containing 117 cases (56 normal cases, and 61 cases with masses) is also available. Click here for more information.

VOLUME NORMAL_01 HAS BEEN REMOVED FROM FTP BUT WILL STILL BE AVAILABLE ON TAPE BY REQUEST. VOLUME NORMAL_02 AND VOLUME SPECIAL_01 ARE NOW AVAILABLE ON FTP AND ON TAPE.

To place yourself on an electronic mailing list to receive updates about this project (including the eventual creation of mailing list discussion group),
Click Here Mail Me Email: ddsm@bigpine.csee.usf.edu


o Nijmegen Database:

All 40 images in this dataset show one or more clusters of microcalcifications. The mammograms are from 21 different patients. Both benign and malignant cases are included. (Benign cases: 1,2,7,8,9,10,14). The position and size of the microcalcification clusters were marked by two expert radiologists, based on all patient data available (different views, magnifications). These annotations were entered into the computer as circular areas and are stored in separate files (extension mrk). Pixel level output files of the microcalcification detection program on the public database of 40 mammographic images are included. This directory also contain files that give the lookup tables for conversion of the image files using adaptive noise equalization The 40 images in this dataset have been made available by the Department of Radiology, University Hospital Nijmegen, PO box 9101, The Netherlands. Questions can be addressed to Dr. Nico Karssemeijer (e-mail nico@mbfys.kun.nl) Publications making use of this dataset should mention this using the following acknowledgement: 'Images were provided by courtesy of the National Expert and Training Centre for Breast Cancer Screening and the Department of Radiology at the University of Nijmegen, the Netherlands'. These images are also available via anonymous ftp to figment.csee.usf.edu.
o Click for information on the images and ground truth files
o Click for information on Nico's detection results and noise equalization look-up tables
o Click for access to the Nijmegen Database


o LLNL/UCSF Database:

We, at Lawrence Livermore National Laboratories (LLNL), along with University of California at San Fransisco (UCSF) Radiology Dept. have developed a 12 volume CD Library of digitized mammograms featuring microcalcifications. For each digitized film image, we have 2 associated "truth" images (full size binary images) that show:
Along with the "truth" images, we provide a file with case history, radiologists comments, and other information.

The library contains 198 films from 50 patients (4 views per patient, but only 2 views from one mastectomy case), selected so as to span a range of cases of interest. These films were digitized to 35 microns. Each pixel was sampled to 12 bits of grayscale. As a result, each digitized mammogram results in image that is about 50 megabytes in size, for a total of nearly 6 gigabytes for the entire library.

The films were selected to present 5 normal, average, healthy cases (previous normal mammograms and no history of ultrasound, magnification views, biopsy, etc.), 5 normal but difficult cases (with either dense or fibrous breasts, implants, or asymmetric tissue), 20 cases of obviously benign microcalcifications (with at least 3 years of follow-up without change or developing cancer), 12 cases of suspicious, benign microcalcifications, (note: all these benign cases had either a biopsy or a diagnostic mammogram plus at least 3 years of subsequent follow-up without change or developing cancer), and 8 cases with a malignant cluster of microcalcifications, biopsy proven.

The set is available for US $100 to cover our reproduction costs. For more information on the database, send e-mail to mammo-db-help@llnl.gov


o Mammographic Image Analysis Society (MIAS) Database:

The images in this database were scanned with a Joyce-Loebl microdensitometer SCANDIG-3, which has a linear response in the optical density range 0-3.2. Each pixel is 8-bits deep and at a resolution of 50um x 50um. Further details of performance can be found in:
The data is in compressed format and before use will have to be uncompressed using the UNIX `uncompress ' command. There is an accompaining License Agreement. MIAS is a nonprofit organization, and there is a small fee for a copy of the database on an 8mm Exabyte tape. Ordering information can be found at the MIAS Website


o Washington University Digital Mammography Database:

This database contains digitally acquired images of pathologically proven breast pathology. Each case consists of a single scout image of a breast lesion obtained during the course of stereotactic core needle biopsy. Cases are organized according to lesion histopathology. Patient demographic data will be added to the database in the near future.

Images have been acquired on a LORAD CCD-based, stereotactic core biopsy table. Each image is 512x512 pixels in size with 12 bits of grey scale data per pixel. Icons of images are available for viewing on the Web page. Raw image data can be downloaded via FTP to any Unix-based system.

Our intent in making these images available is to foster image processing research in digital mammography. To date, most publicly accessible digital mammography databases are made from film digitized mammograms. Image processing on such images does not allow the investigator to take full advantage of the unique characteristics of directly acquired digital data. It is our hope that the availability of digitally acquired mammographic data will aid the further development and acceptance of digital mammography.

Click here to access the Washington University Digital Mammography Database

To contact the database manager: Jonathan Fein -- jafein@artsci.wustl.edu


Mammography Websites:


o The Digital Mammography Home Page at Brandeis University. Contains links to several other digital mammography pages. Includes information on upcoming conferences, digital detector work, computer-aided diagnosis and image processing, and commercial ventures.


Mammography Papers:


o"Registering time sequences of mammograms using a two-dimensional image unwarping technique"

Abstract: The main problem with automatically comparing two corresponding mammogram images is the high level of differences between the two images that do not result from a developing abnormality. These differences are mainly caused by natural changes in the breast, and changes in positioning and amount of pressure applied to the breast from one mammogram session to another. In order to produce meaningful differences by subtracting corresponding mammograms, we must first register the two images by compensating for the changes which are of no diagnostic interest. In this paper we introduce a two-dimensional image unwarping technique as an approximation to reversing the three-dimensional warping which causes the same view of the breast to appear different from one mammogram session to another
Reference:
AUTHOR = "M. Sallam and K. Bowyer",
TITLE = "Registering time sequences of mammograms using a two-dimensional image unwarping technique",
BOOKTITLE = "Second International Workshop on Digital Mammography",
MONTH = "July",
YEAR = "1994",

"Registering time sequences of mammograms using a two-dimensional image unwarping technique" Paper




Please mail comments, suggestions and specific mammography questions to:
ddsm@bigpine.csee.usf.edu